modified kalman filtering method for reducing gps ... · using trajectories.softwares like anfis...

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056

TIME

K.Nithiya1, Mr.A.Vinoth Kannan2, Mr.M.Anantha Kumar3

1 Student,MEApplied Electronics, IFET college of Engineering, Tamilnadu, India 2 Assistant Professor, ECE, IFET college of Engineering, Tamilnadu, India 3 Assistant Professor, ECE, IFET college of Engineering , Tamilnadu, India

---------------------------------------------------------------------***---------------------------------------------------------------------

Tracking a GPS equipped Moving vehicle in Real

time still tends to have Errors while tracking using

user segment Receivers and softwares. This paper

proposes a Different method for tracking a vehicle

moving with frequently changing speed ,using

Genfis1 of ANFIS(Adaptive Neuro-Fuzzy

Inference System) with Matlab 2013a. The

Tracking Accuracy is focussed as important

parameter to be improved than existing method of

RBF based tracking and training method.Kalman

filter is initialized with varied Stepsize and

Covariance is updated in Real time and adjusted to

reduce the Noise covariance of Observed GPS

values.

Keywords: Kalman ,Stepsize, ANFIS, Tracking, GPS, Training. Introduction: Accuracy of GPS Tracking is a major Application in Researchareas nowadays for tracking GPS equipped vehicles, Fleet tracking by companies and spatial analysis using Trajectories.softwares like ANFIS LAB, MATLAB with ANFIS is used to Train coordinates values obtained from accelerometers and GPS receivers.The Modified Kalman Filtering method Combined with ANFIS is used here to Train and track GPS latitude and longitude values from BU353 WAAS enabled GPS receiver.Kalman filter is more effective than Particle filter in terms of computational complexity.

Analysis of proposed method:

the training ANFIS discard the major Disadvantages of

neural and ANN , RBF network by inserting Prior Knowledge as Fuzzy Rules into the Neural network. ANFIS

generates Fuzzy rules by identifying the Membership Functions and optimize error by blending the leraning Rule. Here the GPS trajectory is analysed upon 4 steps

1. Preprocessing stage

2. ANFIS Training Process

3. Kalman Filtering Initialization with Varied step size

4. Covariance Update and adjustment in Real time.

5. Observation of RMSE Errors.

Data

Preprocessing

ANFIS

System

DOM

Feature

Extraction

covariance update

PROPOSED TRACKING METHOD

Pythagoras

Positioning

Observed valueLoad

Input

Data

Membership Function,

Type,Structure

Kalman

Filter

positioning interval,

Initial Velocity,

Initialize Kalman

Filter

Load training data

Constraint

Prediction

Extract

Positioning

Features

Fig 1 - Proposed method based on ANFIS

PREPROCESSING METHOD: Pythagoras method is here used as preprocessing method to find differnce between 2 latitude longitude points.The distance in meters along is displayed as result of preprocessing method This is used as input Estimated values for Error tracking in ANFIS training after defining the membership functions as Generalized Bell and defining Fuzzy rules. Since Pythagorus method tends to have less errors than great circle as method.

Volume: 02 Issue: 02| May -2015 www.irjet.net p-ISSN: 2395-0072

Abstract

© 2015, IRJET.NET- All Rights Reserve Page 32

VEHICLE TRAJECTORY TRACKING ERROR USING ANFIS IN REAL

MODIFIED KALMAN FILTERING METHOD FOR REDUCING GPS -


Volume: 02 Issue: 01| April -2015 www.irjet.net p-ISSN: 2395-0072

OBSERVATION OF PREPROCESSING METHOD

Fig 2 Pythagoras based Positioning

DOM FEATURE EXTRACTION:

Degree of Mismatch feature extraction helps to extract the essential contents of dataset, required ranges,conceptual data from any database. Here excel is used to contain raw data values obtained from BU353 GPS user segment receiver.Raw data evolves with error and missing values are analyzed and the required values such as Latitude and Longitude is extracted. Then cartesian coordinates of orbit parameters are converted to WGS84 datum indication direction with meters.True Trajectory vectors are extracted for measurements of confidence value analysis. True value vs estimated values of kalman filter was formulated by this DOM feature vectors

EXTRACTED PARAMETERS

POSITION FOR

SPATIAL

PROCESSING

TIME

DIFFERENCE

DISTANCE IN

METERS

1.8070e+006 -1.1602e+007 3.1806e+05E

8.6510e+006 -2.0853e+007 3.6432e+006E

1.2966e+007 -1.4356e+007 3.4063e+006N

1.4061e+007 7.4752e+007 4.8070e+006N

1.8059e+006 1.1603e+006 5.1456e+006E

8.6484e+006 -2.0855e+007 5.6722e+006E

1.2968e+007 1.8811e+oo6 1.2345e+006E

1.4058e+007 2.0736e+007 1.18974e+006E

1.8048e+006 1.6543e+007 2.4084e+006E

8.6458e+006 2.1932e+007 2.8745e+006E

Table 1 Extracted Features

KALMAN FILTER INITIALIZATION AND

ESTIMATION OF FUTURE VALUES:

Kalman Filter is linear quardratic estimator here used

mainly for future location prediction using previous values

with inaccuracies.HereTracking a vehicle, Kalman filter

projects Extrapolation of 20 seconds projection into the

future and estimates the future values with respect to past

trajectory values.

The Kalman filter can be thought of as operating in two

distinct phases: predict and update. In the prediction

phase, the vehicle's old position will be modified according

to the physical laws of motion (the dynamic or "state

transition" model).Vehicle can be equipped with a GPS

unit that provides an estimate of the position within a few

meters.

In addition, since vehicle is expected to follow the laws of physics, its position can also be estimated by integrating its velocity over time, Ideally,ifkalman cannot drift away from the real position in case of sudden change in velocity due to the ANFIS training and updating the covariance values , the GPS measurement should pull the

position estimate back towards the real position but not disturb it to the point of becoming rapidly changing and noisy.

Fig 3 Velocity analysis in matlab 2013a

ANFIS TRAINING AND TESTING:

The Neuro-Fuzzy overcomes the main disadvantages of

Neural network and it was a rare platform to be used to

insert our prior Knowledge as Fuzzy Rules into neural

network.In this concept 3 fuzzy rules is used to update the

covariance of GPS data and kalman residual measurement




by training using Hybrid method and evaluating the input

output modelling of trained output vs ANFIS output

Fig 4 ANFIS Training input values

Rules include covariance values with positive negative values and state variation is proportional as the system evolves for 20 seconds.Training and testing process was completed in 1.6 seconds.

Fig 5 Viewing the Trained values in ruleviewer,Matlab 2013a

Totally 60 datapairs and 200 epoch was used.stepsize is an array of scalar values with RMSE errors of training data pairs returned after every epoch.

UPDATING COVARIANCE

The Estimated values from kalman filter not only will a new position estimate be calculated, but a new covariance will be calculated as well. Perhaps the covariance is proportional to the speed of the vehicle because we are more uncertain about the accuracy,position estimate at high speeds but very certain about the position estimate when moving slowly. Next, in the update phase, a measurement of the truck's position is taken from the GPS unit. Along with this measurement comes some amount of uncertainty, and its covariance relative to that of the prediction from the previous phase determines how much the new measurement will affect the updated prediction.

Fig 6 ANFIS evaluation Graph

Below is the GUI design that includes all the processes such as preprocessing, spatial data extraction, ANFIS training and updating covariance.

Fig 7 GUI design including all processess

RESULT: The outcome of ANFIS module results in reduced Tracking error when inferring the updated covariance values.The Tracking error was Reduced upto 0.000139*10^4 when averaging all the 60 values. The fuzzy rules helps to infer

© 2015, IRJET.NET- All Rights Reserve 4



the convergence values of extrapolation in meters/seconds for nearly 20 seconds into future.Accuracy is improved than Existing RBF based tracking results. Thus the Tracking error was drastically reduced than any Existing methods like RBF based training,Bayesian interpolation, FCM based trajectory prediction etc.

Table 2 Results and comparison

Spatio Temporal Analysis includes existing method which does not considers the tuning parameters and separate method for training such as RBF(Radial Basis Functions network)and Fuzzy based clustering for testing with RMSE values lesser than that of proposed ANFIS based tracking method.Accuracy of tracking is not declined for the desired parameters and accounts for about 89.5% which is greater than existing method.

REFERENCES:

[1.] Xu Y K, Liang X G, Jia X H 2013 Adaptive maneuvering

target state estimation algorithm Systems Engineering and

Electronic 35(2) 250-55

[2] Rigatos G G 2012 Nonlinear Kalman Filters and Particle

Filters for integrated navigation of unmanned aerial vehicles.

Elecronic Systems 60 978-95

[3] Zhangsong Shi, Pixu Zhang, Shen Li, Rui Wang 2012 An

adaptive tracking algorithm of maneuvering target. Advanced

Materials Research: Manufacturing Science and Technology

383-390 2179-83

[4] Ramazan H, Mohammad T, Ali N M 2011 A novel

adaptive fuzzy unscented kalman filtering method.

International Journal of Humanoid Robotics 8(1) 223-43

[5] Rigatos G G, Tzafestas S G 2008 Extended Kalman

Filtering for Fuzzy Modelling and Multi-Sensor Fusion.

Mathematical and Computer Modelling of Dynamical Systems

13 251–66

[6] Escamilla A P J, Mort N 2002 Multi-sensor data fusion

architecture based on adaptive Kalman filters and fuzzy logic

performance assessment. Proceedings of the 5th International

Conference on information Fusion 2154249

AUTHOR(s) PROFILE

Author K.Nithiya completed her B.E ECE in PMU Tanjore, presently in the stage of completion of M.E Project Thesis in IFET college of

engineering,Villupuram,India.Her interests include Wireless technologies and neural networks.

This Project was guided by Assistant Professor,Mr.A.Vinoth Kannan ,ECE dept,IFET college of engineering.His interests include wireless and VLSI based circuit innovations.

Mr.M,Anantha Kumar,ECE Dept,IFET College of

engineering,Villupuram.His Interests include handling

Tracking Method

Training and Testing Method

RMSE

ACCURACY

Spatio Temporal anlaysis

RBF-Training FCM -Testing

1.3657

83%

Constraint Prediction

ANFIS 0.01387 89.5%


Neural based projects.

Co-Author of this research article

modified kalman filtering method for reducing gps ... · using trajectories.softwares like anfis...

Documents